Style Miura版本Miura v1.0 (ID: 253739)

Style LoCon for artist Miura.

The artist has two styles and this model is trained for both of them. The first style requires no keyword. Use keyword "realistic" to get the second style. The same keyword can also be used as negative keyword at low weight if that style is unwanted.

The training images were mostly tall so the model is also better at creating tall images. 512x704 and 512x768 seem to work well.

Weights up to 1.00 work. Use lower weight if the style comes out too strong.

Model was trained on AnythingV5. It will work best when used with that model or models similar to it.

Ignore civitai saying that it is AnythingXL it is wrong.

V1.7:

All images were tagged as "miuran315". But the style seems to work just fine without including that word.

Use "realistic" as negative prompt if that style comes out (0.7 weight is good). It is not always needed.

Training data was 348 images of which 250 were of the "base" style while 98 were of "realistic" style.

Model was trained with DoRa so it won't work on older versions of A1111.

V1:

Training data was 242 images of which 148 were of the "base" style while 94 were of "realistic" style.

描述:

训练词语: